ViT-Based Automated Detection of Cervical Intraepithelial Neoplasia in Colposcopy Images

Incorrect diagnoses of Cervical Intraepithelial Neoplasia (CIN) directly impact the increase in the mortality rate from cervical cancer. In recent years, Latin America has been among the regions with the highest incidence and mortality rates. Some works have focused on its prevention, aiming for early diagnosis and follow-up of its precursor lesion, Cervical Intraepithelial Neoplasia, also known as Cervical Dysplasia. Therefore, methodologies based on computer vision and machine learning are vital for developing early diagnostic assistance tools to support specialists. This work reports a novel method based on Vision Transformers to classify the progression grades of Cervical Intraepithelial Neoplasia using colposcopy images obtained from the open database generated for the Intel & Mobile ODT Cervical Cancer Screening Challenge.